Using Data from Satellite Telemetry and Nesting Beach Monitoring to Improve Estimates of Marine Turtle Clutch Frequency

Population abundance data is often used to dene species’ conservation status. Abundance of marine turtles is typically determined using information from nesting beach monitoring such as clutch frequency and remigration interval of nesting females. However, studies have shown that clutch frequency determined solely from nesting beach monitoring data can be underestimated. To obtain reliable estimates of clutch frequency for hawksbill turtles in northeastern Brazil (-6.273356 S., -35.036271 W), the region with the highest nesting density in the South Atlantic, data from beach monitoring and satellite telemetry were combined from 2014 to 2019. Beach monitoring data indicated the date of rst nesting event, whilst state space modelling of satellite telemetry data indicated the departure date of turtles, allowing calculations of residence length at breeding site and therefore estimation of clutch frequency based on internesting intervals. Nesting females were estimated to nest up to six times within the nesting season and had an average clutch frequency of 4.7 nests per female. This estimate is almost twice larger than previous estimates based only on beach monitoring. The new estimates of clutch frequency will allow for more reliable population abundance estimates for this critically endangered population. Further, to guide future estimates of marine turtle clutch frequency methods to estimate clutch frequency were compared and their advantages and biases were discussed. Our approach and ndings highlight the need for reconsideration of how clutch frequency is commonly determined for marine turtle populations and the use of beach monitoring data and satellite telemetry for estimations of clutch frequency.


Introduction
Population abundance data is one of the key parameters used in population assessments to evaluate and determine a species' conservation status (IUCN, 2017a; Willians et al., 2011), where low abundance or decreasing trends may indicate that a population is under threat and may need management intervention (Robins et al. 1999). Ideally, abundance estimates should consider all life stages from individuals within a population and be determined for both sexes (Schwarz and Seber 1999;Iijima 2020). However, this may be di cult to determine, particularly for species with high dispersal, and when different life stages and genders are not equally available for counting (Bradbury et al. 2008). Further, some species may utilize a variety of areas spatially within a region, being residents to core areas, and /or also migrate between locations, being transient in some locations (e.g., often only being observed once at speci c locations), biasing counts and abundance estimates (Chaloupka and Limpus 2002;Clavel et al. 2008). This is the case for marine turtles, where estimates of species abundance should consider information on hatchlings, juveniles, sub-adults and adults from both sexes (Chaloupka and Limpus 2001;Rees et al. 2016). However, determining abundance across life stages is hard, especially for some life stages, such as the rst few years of hatchlings, since limited information is available for this life stage (Putman et al. 2020). Further, until marine turtles reach maturity males and females cannot be distinguished visually, making it di cult to determine the gender of individuals and determine numbers for each sex individually (Meylan et al. 2011). Males are also more di cult to encounter and count as they remain in the oceanic environment for their entire life, and do not come ashore as nesting female turtles, with exception of cold-water regions, where they come ashore to bask, such as in the Galapagos (Maxwell et al. 2014) and Hawaii (Whittow and Balazs 1982). Additionally, marine turtles use a variety of areas across each of their life-stages, having residence in some locations (e.g., foraging areas) but being transient in others (e.g., migratory corridors) (Chaloupka and Limpus 2002; Prince and Chaloupka 2012).
As a result of these factors the easiest, least expensive and most used method to estimate marine turtle abundance is to utilize counts of the number of nests and breeding females at nesting beaches (National Research Council, 2010; Mazaris et al., 2017). The fact that breeding individuals within a population aggregate in particular areas for reproduction makes it easier to count individuals that may disperse across the vast marine realm ). Marine turtles lay several clutches within a nesting season, with intervals between seasons from one to several years, these remigration interval (RI) varies amongst individuals, populations and species (Miller 1997;Troëng and Chaloupka 2007;Warden et al. 2017). A snapshot of the number of females nesting in a given season within a population, can be estimated using the total number of nests across all nesting sites used by that population in that season divided by the number of nests that individuals laid (clutch frequency) (Johnson and Ehrhart 1996;Broderick et al. 2002). However, this approach is simplistic and doesn't account for the uncertainty in this reproductive rate parameter (Ceriani et al. 2019 , 2009). This is the case for hawksbill turtles (Eretmochelys imbricata) nesting in the South Atlantic, in particular in Brazil which has the largest hawksbill turtle nesting density in the region Santos et al. 2013). Thus, to obtain more accurate estimation of population abundance of hawksbill turtles nesting in Brazil there is the need to revisit clutch frequency numbers. Here, we combined beach monitoring data with satellite telemetry data to reassess clutch frequency for hawksbill turtles that nest in the southern coastline of Rio Grande do Norte, Brazil, an important region for this critically endangered species. Further, we provide comparisons among several methods to estimate clutch frequency and discuss the advantages and biases from data collected from each approach to inform future studies that aim to estimate marine turtle clutch frequency.

Study site and approach
Clutch frequency was determined by combining information from nesting beach monitoring and from satellite telemetry obtained at three hawksbill turtle nesting beaches (Chapadão, Minas and Sibauma,

Study design
Morning beach monitoring for marine turtle activity (e.g., nests and false crawls) was conducted daily from November 1 st to May 30 th during the 2014/2015 and the 2018/2019 nesting season. Each record was identi ed as a nest or false crawl, but for this study we only considered information from nests. As well as morning beach monitoring, during each of these ve nesting seasons, intensive monitoring was conducted nightly from the 10 th December to 15 th April, which accounts for 93.4% of nesting across the season. Sunrise in the region occurs around 4:30 am and 81% of nesting records occurs before midnight (Nakamura et al. 2019), thus night monitoring consisted of the beaches being patrolled by at least two people from 7 pm to 3 am to maximize the probability of encountering nesting turtles in our study sites. When encountered, hawksbill turtles were intercepted after egg-laying, checked for Inconel tags (number 681 National Band and Tag Company), and if none were present, they were tagged in both front ippers.
Curved carapace length (CCL ± 0.1, cm) was measured with a exible tape from notch to tip (Marcovaldi and Marcovaldi 1999). After the 15 th April, night monitoring was conducted based on the internesting interval (15 ± 1.5 days) of turtles that had already been encountered laying at our study site (Santos et al., 2010. Although 98% of nests in this region are of hawksbill turtles ) all the nests were excavated after hatching, to con rm the species for the nesting event.
Nesting females encountered during the night monitoring were randomly selected for attachment of platform transmitting terminals (PTT; N = 35), and 10 of them were tracked in two consecutive nesting seasons with a new PTT (Table S1). Six PTT models were used for our study (Table S1; mass in g; dimensions [length x width x height]): 12 SPOT-293A -119 g -72x54x24 mm, 6 SPOT-375 -136 g -99x55x21 mm, 8 SPLASH10-F-296A -195 g -86x85x29 mm, 3 SPLASH10-F-334 -450 g -112x63x62 mm, manufactured by Wildlife Computers (Redmond, WA, USA), and 4 Kiwi-Sat K2G 376D -240 g -136x44x59 mm, 2 Kiwi-Sat K2G 376D -360 g -138x78x50 mm manufactured by Sirtrack (Havelock North, New Zealand). Turtles selected for PTT attachment were restrained in a wood box as per Hart et al. (2010). The carapace was sanded and cleaned with isopropyl alcohol, and PTT was attached with epoxy, followed by anti-fouling paint following protocols by (Santos et al., In press). Turtles in the study site take at least 40 minutes to successfully nest, therefore FastGPS were set to enter haul-out cycle after 20 minutes when the wet/dry sensor was dry and exited haul-out after 30 seconds if wet in the rst nesting season (2014/2015). For the other nesting seasons (2015/2016 to 2018/2019) haul-out cycle was de ned after 5 minutes to make it more sensitive to nesting attempts. If the turtle remained out of the water haul-out messages were sent following the settings of the prede ned Fastloc sampling interval, ranging from one to four locations per hour, enabling the detection of nesting attempts. In cases where subsequent haul-out were detected, i.e., in consecutive days or even in the same night, the last haul-out position was used to represent the nesting event.
For individual clutch frequency estimation, we selected a subset (N = 18) of satellite tracked females that nesting monitoring data indicated laid their rst clutch during the rst portion of the nesting season (Fig.  S1). We selected turtles that nested by the end of January as indicated by data from the monitoring activities, to avoid considering turtles that nested previously during the nesting season ( Fig. S1), which represents 24% of the clutches laid (Fig. S1). PTT deployment for the selected turtles identi ed to have nested in January, took place in December/January (N = 8) February and April (N = 10; Table 1; Fig.   2). Among the subset of tracked females, 11 PTTs (Kiwi-Sat and SPOT models) relied on ARGOS positions while the 7 SPLASH models included FastGPS and haul-out data ( Table 1).

Clutch frequency estimates
Observed clutch frequency estimates (OCF) was based on information obtained during nesting beach monitoring and determined as the number of times that each individual nesting turtle was encountered nesting successfully during the nesting season throughout our monitoring activities.
For individual turtles for which the interval between nesting encounters, during our nesting monitoring, was greater than the range of a typical internesting interval (12 -20 days; Santos et al., 2013), we assumed that turtles nested in that period and calculated an estimated clutch frequency (ECF b ) based on the beach monitoring data. For example, if we had an interval of 30 days between the encounters of a turtle, we assumed that one unobserved nesting event occurred, for intervals of 45 days we assumed that two unobserved nesting events occurred (Johnson and Ehrhart 1996;Broderick et al. 2002;Briane et al. 2007; Santos et al. 2013). In order to investigate whether or not the assumed nesting event occurred whithin the study site, we analyzed all the nesting records that occurred between 12 and 20 days after the last nest in which the individual turtle was encountered within our site. If all the nesting records at our study site during the relevant period were from other identi ed individual turtles it meant that the assumed nest occurred elsewhere outside of our study site. However, if occurred nesting events existed where the individual turtle was not encountered/observed (i.e. nests missed by patrolling personel and encountered by the track), it is possible that the nest was from the turtle in consideration, and therefore we assumed that the nest occurred at our study site.
To estimate clutch frequency (ECF s ) using the data from the subset of satellite tagged turtles and from nesting beach monitoring the length of residency at the breeding site was divided by the average internesting interval after the rst clutch was laid as suggested by Esteban et al. (2017). Hence the individual residency length (IRL s ), was rst determined by tting a hierarchical state-space modeling (SSM) (Jonsen et al. 2006) using location estimates for each 6 hour period from the satellite tags to determine when the turtles departed from the breeding site and started migration to their foraging areas. As the turtles were instrumented during the nesting season, the ARS behavior before 'transiting' behavior (migration) was linked to internesting and therefore used to indicate the residence period at the breeding site. The model that converged better was based on 40,000 iterations after a burn-in of 60,000 samples and thinned by ten to minimize within chain sample autocorrelation. We further excluded the data for migration and foraging as it was not relevant for this study. In the ten cases where the turtle was detected nesting during beach monitoring prior to satellite tag instrumentation, we added the number of days since it was rst recorded nesting within that season to the calculation of the IRL s at the breeding site. Similarly, two females (ID 13 and 25) started migration immediately after instrumentation, however, they were detected by the nesting beach monitoring when they nested earlier in the season allowing us to estimate their residence period at the breeding site (Fig. 2). For the cases (N = 3) that SSM did not detect behavioral changes associated with migration (i.e. local turtles), we inspected their home range contours using OpenJUMP HoRAE program (Steiniger and Hunter 2013). The 95% contour from Scaled Line-based Kernel Density for Movement Points function was used and the split in the home range output allowed us to identify the exact departure day when the turtle left the breeding site ( rst polygon) towards its foraging site (second polygon), allowing us to determine their residency length (Fig. S2).
For comparison purposes, we also calculated the individual residence (IRL b ) length based on beach monitoring data, as the difference in days between the last and rst nests recorded for all individuals (N = 122), excluding the transient turtles, that is, those observed only once (N = 88). We further compared IRL b as well as the number of transient turtles from the rst portion of the nesting season with those that started to nest later in the season, that is, with rst nesting from February onwards. In addition, we calculated population residence length (PRL) at the breeding site by determining the average day for the rst nesting, based on information from nesting monitoring (N = 210), and the average departure date based on SSM (N = 32) or home range inspection (N = 3; IDs 13, 16 and 16*). December 1 st was considered day zero for each nesting season since nesting for this populations typically starts early in December . The difference in days between averages of rst nesting and departure were used to estimate PRL at the breeding site. Similarly, to the ECF s calculation described above, we used PRL to obtain a clutch frequency estimate for the nesting population (ECF PRL ) dividing PRL by the average internesting interval and adding 1 clutch to take into account the rst nest.
To further inform future studies that aim to estimate marine turtle clutch frequency, we compared the ECF b based on beach monitoring solely with ECF s based on the combination of beach monitoring data with satellite telemetry using a paired t-test and determined the number of unobserved nests by beach monitoring. Similarly, we compared residence length that incorporated data from satellite telemetry (IRL s ) with those obtained from beach monitoring solely (IRL b ).
We compared IRL s and ECF s for the individual turtles tracked in two subsequent nesting seasons, however we did not make any comparative statistics as the sample size was small (N = 4). We also evaluated the e ciency of satellite telemetry in determining nesting events based on the haul-out data in comparison to the nesting records by beach monitoring and length of residence period at the breeding site.

Nesting activity
Between 33 to 51 (average 42 ± 7.5) individual nesting hawksbill turtles were encountered each season during the beach monitoring conducted at our study site, with in average 124 ± 8 (range 114 -138) nests per season (Table S2). Of the individuals encountered during the beach monitoring, 41% were transient and seen nesting only once during the nesting season within our study site (Table S3). The average IRL s for the subset of tracked hawksbill turtles at the breeding site was 55.9 ± 11.8 days (range 31 -76 days; Table 1; Fig. 2), whereas the PRL at the breeding site was estimated to be 52.6 days (

Assessment of information obtained from nesting beach monitoring and satellite telemetry
Nesting females were encountered by the monitoring team for most (82.6%) of the nests recorded in our study during our monitoring activities (Table S2). When calculating the ECF b (Table 1), 11 nests were assumed to have occurred based on internesting interval (Table S5). Of those nests, three were assumed to occur outside of our study site, as there were no missed records by eld personnel during the probable internesting period (i.e., records in which the female was not observed) (Table S6). Eight nests were assumed to have been missed by beach monitoring, since the females were not encountered during the monitoring patrols, but tracks were recorded during the relevant period that they were likely to nest (Table  S6).
Satellite telemetry information typically complemented nesting data from our monitoring efforts, for instance three individuals (IDs 8, 19 and 20) were observed nesting only once during our beach monitoring (Table S5). However, satellite telemetry data indicated that residence length at the breeding site was compatible with at least three other nesting events at our study sites during the nesting season (Fig. 2). Further, we assumed that three nests occurred outside our study site; however, information from haul-outs obtained from the FastGPS tags indicate that one of them actually occurred at our study site but was missed by the beach monitoring (Table S6; A7). Nevertheless, two nesting events observed during the nesting monitoring were not recorded or transmitted by information obtained from haul-out from one of the satellite tags (ID 9*; Table S5). Another two nests were assumed to be not recorded or transmitted by haul-outs; however, they were not con rmed by information from beach monitoring, since the period that individuals (ID 3* and 4*) remained at the breeding site (14 and 16 days) indicates that they probably nested but that it was not detected.

Discussion
Combined data from nesting beach monitoring and satellite telemetry allowed us to provide robust estimates of average clutch frequency for hawksbills turtles nesting in Rio Grande do Norte state, Brazil, with estimates ranging from 3.3 when using observed data from beach monitoring to 4.7 when combining the beach monitoring data with satellite telemetry. Larger clutch frequency estimations when using satellite telemetry and data from nesting turtles opposed to only using data from nesting beach is unbiased clutch frequency estimates with beach monitoring data solely. Clutch frequency using satellite telemetry has been calculated for nesting hawksbill turtles in the Dominican Republic (between 2-4 clutches); however, the selection of individuals for these estimations may have included females that nested previously in the season, underestimating clutch frequencies (Revuelta et al. 2015).
The temporal scale of monitoring, and consequently data inclusion into estimates, affects clutch frequency estimates; thus, care should be taken to ensure that the whole nesting season is incorporated into such estimations. Bio-logging tools such as radio or satellite telemetry are very helpful to keep track of internesting returns ( (Goldberg et al. 2013). Lastly, the process of PTT attachment may in uence nesting turtle behavior, in particular if the turtle is displaced to another area for the instrumentation and further release, which was not the case for this study (see Luschi et al., 2003Luschi et al., , 1996. Our estimates for residence length that incorporated data from satellite telemetry were close to each other IRL s of 55.9 ± 11.8 days (range 31-76 days; N = 18) and the PRL of 52.6. It is interesting to note that this single number (PRL) was estimated based on the difference between average date for rst nesting event from beach monitoring (N = 210), which includes turtles seen only once (transients), and the average date from departure from satellite telemetry (N = 35) (Fig. S4).  ) found that hawksbill turtles that arrive earlier in the nesting season occupy shallower waters, which may be associated with higher quality breeding residence habitats. In this sense, the arrival time for the breeding season may also in uence residence selection, which possibly affects its length and therefore clutch frequency estimates. One way to investigate the possible in uence of arrival timing during the nesting season on clutch frequency would be to deploy satellite transmitters before the females arrive at the breeding site. This would require the device to function for more than two years, as the remigration interval for hawksbill turtles is typically two years  or to attach the equipment in the foraging ground prior to migration (see Pilcher et al., 2020). However, selecting turtles at foraging grounds that will likely start migration to breeding areas is challenging since it would require the identi cation of individuals that are reproductively ready to leave, this could be determined with laparoscopy or ultrasound (Pilcher et al. 2020). It is speculated that the longer the distance from the foraging ground to the breeding site the more energy turtles will spend on migration (Enstipp et al. 2016) and therefore less energy may be allocated to reproduction, resulting in smaller clutch frequency or clutch sizes (Patel et al. 2015 likely to in uence the quality of foraging habitats over time (Hays 2000), there is the need to revisit demographic parameters such as clutch frequency and remigration interval from time to time.
The spatial extent of sampling, and data inclusion, also needs to be considered for clutch frequency estimation. It is very common for projects to report their nest count for their speci c study site  Table S2) that also used adjacent beaches to nest ). There are studies with hawksbill turtles that exclude transient turtles for the calculation of clutch frequency (Beggs et al. 2007). Using this approach for the previous example, the estimation would result in 3.9 clutches per female, which is the same ECF b that we obtained; however, if we just divide the total number of clutches by the number of individual females observed, it would result in 2.9 clutches per female. Therefore, sampling a fraction of the population will always require dealing with biases from transient turtles. Sometimes turtles are wrongly considered transient as a re ection of low detectability by beach monitoring (Pfaller et al. 2013). For example, in our study even with an intensive monitoring effort, nests were missed, as was evidenced by the haul-out from female ID 4* (see Tables S5 and S6). Tidal regime may play an important role on detectability, especially during high spring tides, as turtle tracks can be erased by waves. Indeed, the missed record A7 (Table S6) occurred during spring tide. In our study site, hawksbill turtles often (48%) nest below the highest spring tide line, as they crawl up the maximum possible path and come across a sand slope that is exposed to high spring tides (Santos et al. 2016). Thus, tidal regime is an important factor to consider when designing monitoring surveys and interpreting nesting information. Additionally, despite the fact that most hawksbills turtles nest at night, a few nests Our results indicate that beach monitoring data combined with satellite telemetry information may ne tune clutch frequency estimation for marine turtles. However, it is important to consider that if the internesting habitat for hawksbill turtles is very close to the shoreline (e.g. less than 1 km), detecting nesting events through satellite telemetry will require tags with ne scale resolution, which are usually costly (Esteban et al. 2017). Thus it is suggested that residence length should be used as indicator of clutch frequency, such approach would allow for sample sizes to be higher through the use of lower resolution cheaper satellite tags. Our approach and suggestions although applied to hawksbill turtles nesting in Rio Grande do Norte, should be considered for other nesting locations in Brazil (e.g., Bahia, Piauí, Paraiba, Pernambuco, Alagoas and Sergipe states), which would allow for nesting numbers to be estimated for this endangered genetically discrete population.

Declarations
Acknowledgements The satellite telemetry data for this study was acquired as part of a licensing requirement for seismic surveys carried out by Petroleum Geo-Services (PGS) and Spectrum Geo. Field work was conducted by Pro-TAMAR Foundation with participation of Centro TAMAR-ICMBio. We are thankful to Santuário Ecológico de Pipa, Paulo Barata, Paolo Casale, Erik Allan Pinheiro dos Santos who helped with this work and manusript and all eld workers that helped on during beach monitoring activities.
Compliance with Ethical Standards Biodiversity authorization and information system (SISBio) issued the data collection license 42760, respecting Brazilian animal care regulations. All authors declare no competing interests.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryInformation.docx